3.0b Release:This latest release combines my two checkpoint merges into one, along with a couple other models and some QoL LoRAs to give consistently better results. Unlike Cutie, I'm pushing this one further in the anime direction. By default, this model will produce more cartoonish/anime images than prior versions, but can easily be coaxed into making other styles. This update should give more consistent results with hands and tails than previous versions. One downside, though, is that sometimes tails can go a little crazy. This can generally be fixed by putting in a tail tag at a low strength (i.e. (tail:0.2)).Recommended SettingsSampler: Euler a / DPM++ 2M KarrasCFG: 8-13Steps: 40+CLIP Skip: 1Notes: This new version responds very strongly to embeddings and TIs. Please try to create your image without those first, and then only use them sparingly for best results.Initial Release:Personal checkpoint merge of CoconutFurryMix2.0.Av1_Vivid and IndigoFurryMixvX_AnimeWent through a few iterations before the mix turned out right. Seems to consistently give me closer to my desired result than either of them. Figured I'd upload it so Civit would stop saying there were no associated resources for the images I made with it. Works well with both SFW and NSFW.Recommended Settings:Sampler: DPM++ 2M Karras / DPM++ 3M SDE KarrasCFG: 6 - 13Steps: 20 - 60For best results, use of the ADetailer extension is recommended.Version 2.0 and higher seems to give better results without ADetailer.Lastly, while the model works fine with short prompts, it seems to perform best with longer ones. The more specific you are, the better your image will be.Samples:Caveats:I've noticed that, with some species, Coconut and Indigo have wildly different ideas of how to portray them, and this can lead to some funky images. For these, I'll usually find a good one within 5 generations.
